OCCUPATIONAL STRUCTURE IN ENGLAND AND WALES
DURING THE INDUSTRIAL REVOLUTION
BY
JAN ERIK MEIDELL
PhD THESIS
Department of Economics
I would like to thank Liam Brunt for his unlimited patience, endless energy and passionate interest in the subject. Without his support, this thesis would never have been finalized.
Contents
Introduction 4
1. Trade directories as a data source on occupational 14 structure: evidence from England in 1851
2. How fast, and how broad, was British industrialization? 38 Evidence from a synthetic occupational census for 1801
3. Regional specialization in England and Wales in 1801 and 1851 111
Introduction
0. Introduction. The “industrial revolution” designates a process starting in the mid- eighteenth century, during which England and Wales experienced significant transitions in manufacturing and manufacturing processes and strong economic growth. Helped by a number of new technical inventions – in addition to strong population growth, favourable conditions for commerce, government policies and financial innovations – efficiency in the textile and iron industries increased manifold. Manufacturing moved away from handmade goods and cottage industries into mass production, helping Britain to become a world power. Urbanization changed the face of the country and improvements in agriculture allowed to the growing population to be supplied with food. The industrial revolution has thus had a major historical impact and is seen as crossing a threshold into modern economic growth. It has changed our lives, life expectancy, employment and how we look upon the world. The three chapters of this thesis attempt to better measure the occupational changes in the English and Welsh population during this time period – one of the key characteristics of the industrial revolution.
1. Literature. The Industrial Revolution is well researched; the causes and processes have been greatly discussed, in addition to various analyses of basic national income. An important branch of research has focused on the strength of economic growth, which recent research paints as being slower than originally supposed. Moreover, revisionist social historians assert that “English society before 1832 did not experience an industrial revolution let alone an Industrial Revolution “ (Jonathan Clark, 1986, pp. 39). During the time from 1750 to 1850 the number of patents grew fast as did cotton output; but national income and aggregate consumption grew only gradually.
So national accounts and various local analyses are apparently not enough to answer questions relating to changes in the economy and, in particular, in social structures. In the words of Joel Mokyr (1993, pp. 2), “arguments about what exactly changed, when it started, when it ended, and where to place the emphasis keep raging”.
More recent work looks away from national aggregates to focus on regional and industrial aspects of the industrial revolution, emphasizing acceleration in efficiency and output. But again, the main focus is on the sorts of products, their quantity and the production processes. There is so much more to this time period. How did the industrial revolution impact professional and social structures? Do service occupations expand and, if so, in which areas? Which secondary industries employed the bulk of the workforce, and where? What would a national occupational map of this period reveal?
Mokyr (1993) divides research about the industrial revolution into four schools:
social change, industrial organization, macroeconomics and technology. Of particular interest is the research on macroeconomics. Important proponents of this school are Deane and Cole (1969), with their estimates of national income and industrial production, as well as the more recent revisionists to economic growth – such as Wrigley (1987) and Crafts and Harley (1985). Crafts and Harley claim that productivity growth was confined
to only a few sectors, such as wool, cotton, iron and machinery, with the remaining sectors experiencing much lower productivity increase. Temin (1997) refutes this view, claiming that foreign trade figures show that Britain remained competitive not only in these industries but also in a range of older industries such as linen, glass, buttons, soap and brewing, amongst others. Thus, claims Temin, efficiency must have improved on numerous fronts. Mokyr (1993) holds that the aggregate effect of the Industrial Revolution before 1820 is not significant but has no numbers to support the claim. He further divides the pace of the economy into two groups: traditional trades such as agriculture, construction, craftsmen and domestic industry; and the modern sectors such as cotton, iron, smelting, chemistry, mining and engineering. The missing piece of the puzzle, however, in the preceding discussion is any quantification of professional occupations and social structures. This is a pre-requisite to understanding how the economy and society were impacted by the fundamental structural changes that occurred.
Filling this gap is the aim of this thesis, in the form of creating a national and regional synthetic census for 1801.
From 1801, a decadal national census was taken in Britain. Early versions were rudimentary and the first census with a moderate level of detail was that of 1841. The 1851 census is of superior quality and is frequently used by economic historians as a source to quantify social and professional structures. Comprehensive national and regional tables, with the same structure as the 1851 census, do not exist prior to 1851.
However several attempts were made at quantifying occupations. Patrick Colquhoun (1806) approached the task in the midst of the industrial revolution and created social tables using the 1801 census. These incorporate only a few professions for certain regions, and thus suffer from severe limitations and are of limited use for historians.
Lindert (1980) introduced a new approach, using parish registers to quantify occupational structures in the eighteenth and early nineteenth centuries. Even though these are not representative of the population because they are biased (only male professions for certain age groups and for certain congregations are recorded) they nevertheless represent a very informative source compared to the few data available until then. This line of research has been continued recently by the Cambridge Group for the History of Population and Social Structure, led by Wrigley and Shaw-Taylor, which has constructed regional and national tables based on extended sampling of parish registers and militia ballots from the years 1813 to 1820 (thus dating their benchmark to 1817).
The three chapters of this thesis use sampling from trade directories – in combination with several other sources, such as parliamentary reports and farm surveys – as a data source. My research adds several novelties to the discussion of professional and social structures in the industrial revolution. First, the statistics generated here are for 1801, in the middle of the first industrial revolution and almost two decades earlier than the benchmark generated by other scholars. Second, my data includes both men and women of all ages and congregations and is less biased than the sources used in other studies. Finally, the 1851 regional census structure – comprising 369 professional and social titles divided into 17 groups – has been applied to the 1801 tables. This permits direct comparison to the 1851 census on both national and regional levels; this allows for a quantitative flow analysis during from 1801 to 1851, which is the key time frame for understanding the occupational and social process of industrialization.
Data on the business structure of the private, non-agricultural sector are drawn from the Universal British Directory (UBD), which was published in nine volumes between 1793 and 1798. The UBD was a combined Yellow Pages and White Pages of its time. It offered very extensive lists of tradesmen in each town, as well as separate sections for gentry, clergy, lawyers, doctors, bankers, the town corporation (i.e. town management), substantial outposts of Government (such as Royal dockyards or the Customs Service) and transport (masters of coaches, barges and locally-based ships). In the case of London, the section on tradesmen alone covers 260 pages and amounts to around 34 000 entries; in the case of Manchester, the section on tradesmen covers 72 pages and amounts to around 8 000 entries; and in the case of Birmingham, the section on tradesmen covers 32 pages and amounts to around 3 200 entries. Smaller towns obviously required fewer pages, with the smallest having as few as one page or a half-page. Each entry in the UBD typically recorded the name of the individual (or partnership) and their line of business; in some towns it recorded also the address. It is noteworthy that many individuals and partnerships operated in several lines of business, sometimes up to six, and these were dutifully reported in the UBD. The UBD covers around 1 600 towns and villages across England and Wales, although for many of the smaller towns it does not record details on the businesses that were in operation. Instead, it simply gives a general description of the place and perhaps details on coach connections and such like. We do not know why the details on businesses were reported for some small towns and not others; as far as we are aware, there is no systematic bias.
2. Method. The first chapter of the thesis proposes a method for constructing synthetic 1801 census tables. Using the 1851 census and contemporary trade directories, it can be shown that it is possible to infer local occupational structure using trade directories with acceptable accuracy. The structure of the historical problem is displayed in table 1 below.
The goal is to track occupational change over time using the census but there is no satisfactorily reliable census before 1851. If there is a valid statistical relationship between trade directories from 1851 and the 1851 census, could one use this to create a synthetic census for 1801 by using contemporary trade directories?
Table 1. Data sources available to track occupational change.
1801 1851
? Occupational census
Trade directory Trade directory
Trade directories tell us about the number of businesses operating in each occupation, not the number of workers employed. One approach to construct an occupational structure would be to multiply each business by an employment factor that is appropriate to that occupation. The 1851 census can be interpreted as an enormous and completed trade directory for Great Britain as it contains a table of employees per business, broken down by occupation. Dividing the total number of people in each occupation by the average number of employees per business should therefore give the number of businesses in each occupation. That is, it forms a sort of national trade directory. For the trade directories a similar approach can be followed. Divide the census data by the trade directory and thus create a synthetic employment table. This table is
exact, in the sense that it matches the two series perfectly, by construction, so there is now an employee per business table for 1851, which can be uses to reflate the register of businesses for 1801. For this to be valid, the weights reflected in the tables must be stable between 1801 and 1851, but even though the establishment size should change during this time period, it will not generate a bias as long as it changes equally over the professions. Using this approach does not give an exact number of observed businesses, but I am not trying to find the number of businesses. All I am trying to discover is the distribution of businesses (and, from there, the distribution of individuals’ occupations).
If the trade directory is a random sample of businesses in a particular town, then a one per cent larger share accruing to a particular occupation in the census will be reflected by a one per cent larger share accruing to that occupation in the trade directory (the coefficient on the census data will be unity). To the extent that there is measurement error in the estimated occupational structure derived from the trade directory, the estimated coefficient in the regression should be biased downwards, for standard econometric reasons. Hence I expect to observe estimated coefficients that are less than unity but hopefully not statistically significantly different from it. I entered data from trade directories around 1851 for nine sample towns and undertook the regression. The distributions of the census and trade directories were fairly similar for each town, and the coefficient on the census was not significantly different from unity and with an acceptable r-squared (see chapter two). These results suggest that the 1851 census can generate an occupational distribution of businesses that mirrors that found in trade directories – both at local and national levels. The results also imply that it is safe to work in the other direction – i.e., infer the occupational distribution that we would observe in the census (if this information had been collected) from trade directories around 1801. I feel that these results are satisfactory and act as a “proof of principle”; occupational structure can be inferred from trade directories.
3. 1801 synthetic national census. Chapter two applies the method outlined above to construct a synthetic census for 1801. A stratified sample of towns is taken from the UBD and used to construct estimates of both national and regional business structures, based on the entries for approximately 80 000 individuals operating 100 000 businesses. I then move from business structure to occupational structure using estimates of workers per business establishment. Since the trade directories essentially report only urban data, I supplement these data on industry and services with estimates of agricultural employment (based on the 400 farms surveyed by Arthur Young) and other primary sector occupations (based on various Government enquiries). I also adduce data on the Government sector, which is covered only erratically in trade directories but which turns out to be a crucial consideration. Finally, I estimate the size of the non-working population. The 1801 census provides hard evidence on total population size, so I take 1801 as my benchmark date. Combining all these sources gives a fairly complete picture of the English and Welsh workforce in c. 1801, near the beginning of industrialization.
Hence I refer to it as a ‘synthetic occupational census’. Since the goal is to track temporal changes in occupational structure, I compare my results from 1801 to the census of 1851, near the end of the first industrial revolution. I ensure that the two cross sections are fully comparable by classifying all the workers from 1801 according to the occupational
classification scheme used in the 1851 census, which is generally accepted as the most complete investigation of occupational structure.
The main finding is that industrial employment increased as Crafts-Harley assumed, and much faster than implied by Kitson et al. Industrialization was broad, consistent with Temin’s findings on export growth. In table 2 below I present my results alongside those of Crafts and Kitson et al. The Crafts data have been used repeatedly over the last 25 years as a basis for estimating economic growth; the Kitson et al. results are very recent and have been causing people to rethink the pace of industrialization. My PST distribution is very close to that proposed by Crafts. I have somewhat fewer workers in agriculture, and correspondingly higher shares in industry and services, but the difference is very small. By contrast, the Kitson et al. data show a much higher share of industrial workers already by 1817, and a much lower share of service workers. An important caveat – as Kitson et al. state very clearly in their numerous papers – is that their data pertain to males only. Hence their estimates are not strictly directly comparable to the other estimates in table 2.
Table 2. Comparison of estimates of occupational structure.
1800 (Crafts)
1801 (Brunt-Meidell)
1817 (Kitson et al.)
1851 (Census)
Primary 40 38 38 28
Secondary 30 31 42 41
Tertiary 30 31 19 32
Sources: 1800 – Crafts, British industrialization, 62; 1817 – Kitson et al., “Occupational structure”, 10.
I observe a significant increase in the share of industrial employment between 1801 and 1851, up from 31 to 41 per cent. This is similar to the increase postulated by Crafts and Harley, based on the very imperfect data provided by Massie. The industrial increase is matched by the fall in the agricultural share from 38 to 28 per cent. There was also a very slight increase in the service sector from 31 to 32 per cent. Service sector employment was inflated in 1801 by military mobilization, which accounted for 3.5 per cent of total employment. A counterfactual supposing that military enrolment was only 1.2 per cent of total employment (as in 1817) suggests that industrial employment over the period would have risen from 32 to 41 per cent; services would have risen from 29 to 32 per cent; and agriculture would have fallen from 39 to 28 per cent. This increase in industrial employment is only marginally lower than that supposed by Crafts and Harley.
Overall, the new employment data provide no motivation to revise substantially the existing estimates of economic growth, nor our understanding of the underlying mechanisms that drove them.
One aspect of industrialization that may need to be revised is industrial concentration during the time. I offer no comment on output or productivity growth but I can say that employment growth in cotton and iron was modest. Employment growth in other sectors was much more quantitatively important (apparel, construction, food and beverages). Most interestingly, there were small contributions from virtually all sectors, showing that industrialization was very broad. This lends support to Temin’s analysis of trade data, where he finds that England increased its exports in a wide range of industries.
4. Regional Specialization. The third chapter takes a complete and quantitative approach to regional and sectoral analysis. For each of the 45 English and Welsh counties, a population breakdown using the structure of the 1851 county tables is made. This classifies the population into 369 occupations in 17 occupational groups. The first step in estimating the 1801 county census consisted of fixing the county populations. The 1801 census itself estimated the total population of England and Wales to be 8 872 980, not including military personnel, seamen and convicts. The population is split into counties and this is the starting point for the regional analysis. The next step is to find the number of military personnel, seamen and convicts – which was large in 1801 – and distribute them across the counties. Parliamentary inquiries give detailed information about defence and seamen; for convicts I lean towards Howard’s prison census from 1776. I do not have any reason to believe that the prison population changed significantly to 1801, so the numbers are simply added. The estimates of these occupations sum to 290 087 persons.
Adding these gives a total population of 9 163 067 for England and Wales in 1801, suitably distributed across counties.
The next step consists of deciding how best to estimate the number of persons for each of the 369 occupations in the 1851 county census. Each occupation is attributed to a category based on the data source or method used to estimate the number of workers. Six rules are set up, one for each category. Rule 1 is the simplest. There are five occupations in the 1851 census, which did not exist in 1801, or existed with a negligible number of occupants; these are activities connected with the railway. Hence the estimate for 1801 is nil for each occupation.
Rule 2 incorporates occupations for which I have access to detailed historical sources. Governmental activities (including Customs, Inland Revenue, police and military), fishermen and miners are among the 26 professions listed. The main sources for these are Parliamentary enquiries and reports from 1801 or thereabouts, which give either a detailed list of the incumbents or sufficient information to make an educated guess, both with regards to total employment and the county distribution.
Rule 3 is slightly more complicated. Thirteen of the occupations, and a substantial amount of the population, were in farming. The main source for Rule 3 activities is a survey of 400 farms in c. 1770 prepared by Arthur Young. The survey estimates the ratios of farm servants, agricultural labourers, boys and maids to farmed acreage. For 1801, we have estimates of the total of farmed acreage for each county and we also know the total farming population of England and Wales. The average farm size in 1801 was 146 acres. We use these to estimate the county numbers of four occupations related directly to farming for 1801. For the other nine occupations, which are quantitatively small, we apply the same ratio of farmer to the occupation as given in the national census for England and Wales in 1851.
Rule 4 covers 169 occupations, the largest number of activities, and is thus central to the county distributions. Most of the occupations of historical interest are included here, such as cotton manufacturers, weavers, iron manufacturers and woollen cloth manufacturers, to name just a few. I use the UBD as my main source and use the method discussed above, but on a county level. Thus for each county, I entered data from the 1851 regional (county) censuses as a starting point and pursued a similar approach to the national tables.
Rule 5 contains occupations for which no sources give satisfactory data. Of the 150 occupations present in this group, 64 belong to the category “other” (such as “Other Teachers”). Each group within each class of the 1851 county census has on occupational title of this type. This was an approach chosen by the census authors to reduce the number of occupations from 1091 to 369, by summing up rare occupations for each group within each category. The best I can do for these is to assume that the ratio to the county population in 1801 is the same as the ratio to the county population in 1851.
Obviously, these occupations are numerically unimportant – which is exactly why they were aggregated.
Finally I have five occupations where the number depends on that of another occupation. For example, I take the number of butcher’s wives as a proportion of the number of butchers. For these I apply the same ratio between the two as in the 1851 county census.
When comparing 1801 to the corresponding 1851 census tables, I am able to quantify many of the historical developments found and discussed over the years in the literature on the early industrial revolution. With a notable exception, namely transport, most occupations were more concentrated in 1851 than 1801. For literally all regions, the defence of the country employed fewer people in 1851 than in 1801 (the height of the Napoleonic wars). Although there were relatively fewer farmers in 1851 than in 1801, the counties surrounding London and Lancashire actually became more specialised in farming. The number of mechanics more than doubled and exploded in Lancashire and surrounding counties. Woollen cloth production lost importance and became very heavily concentrated in and around West Yorkshire. Flax and linen manufacture was widespread in 1801, but collapsed before 1851. Cotton manufacture was more labour-intensive but also more widespread in 1801; by 1851 many counties had only negligible production capacity left and Lancashire had gained ground. The overall level of industrial specialisation, on average, remained very stable from 1801 to 1851, but there are important clusters within specific occupations.
Marshall’s (1890) discussion of the concentration of specialized industries into particular localities fits well with the occupational shifts and the regional movements between 1801 and 1851. Marshall noted several causes of the formation of such
“industrial districts”. Primary causes are physical and geographical conditions – such as accessibility by land or water, or closeness to minerals and cheap energy (coal). In addition, for a skilled workforce in one particular sector “mysteries of the trade become no mysteries”, as workers in similar and local industries share their experiences, inventions and improvements both “in machinery, in processes and the general organization of the business”. Marshall further holds that improved transport is crucial, as it allows for splitting up processes and locating different stages of production in different places. It allows, for example, certain counties to specialise in food production and others in manufacturing, whilst allowing both to get their essential inputs (food for factory workers, ploughs for farmers). In general, the observed tendencies between 1801 and 1851 follow Marshall’s assertions. Marshall compared the 1851 and 1881 census, and noted that the population moved away from agricultural occupations into mechanical and manufacturing work, but also tertiary engagements such as education, domestic services and building. Very much the same dynamics can be seen from 1801 to 1851. But the period 1801 to 1851 was marked not only by a rapid increase in the national level of
industrialization, but also by the clustering of several important sectors into Marshallian industrial districts. Woollen cloth production centred in and around West Yorkshire, cotton manufacture in Lancashire, flax and linen manufacture in Lancashire and West Yorkshire, lint manufacture in Lancashire, iron manufacture in Monmouthshire, Cumberland and Worcestershire, lace making in Bedfordshire and Derbyshire, earthenware in Staffordshire, button making and nail making in and around Birmingham.
As one would expect, transportation, on the other hand, became more dispersed.
Ship-agents, boat and bargemen, ship owners and warehousemen settled along the new transport network. Service professions, such as accountants, barristers and lawyers followed new business and settled to a higher degree outside London; so did Government service workers such as police officers and post offices. Engineering and construction professionals neither concentrated nor dispersed; but the number of workers exploded.
Industry needed machines and the workers needed houses. Counties outside the new industrial centres specialized in farming; the counties surrounding London are a good example.
Some of these findings are not new. The concentration of the woollen industry is well documented, for example. But I have found new and reliable sources to quantify this evolution in a detailed and comprehensive manner. Using the UBD, and several other sources, I have been able to construct an 1801 regional occupational census with a satisfactory degree of precision and confidence. The numbers confirm developments highlighted in the literature, and this supports the rigor of our tables. However, there are still numerous open questions to be answered by future research. Thus an internet site is accessible at www.1801census.com with all my data tables and details for those who would like to make further analysis.
Sources.
Allen, Robert C., “Why was the Industrial Revolution British?” in Oxonomics, Volume 4, Issue 1, pages 50-54 (June 2009).
Allen, Robert C., “Agriculture during the Industrial Revolution,” in Floud, Roderick, and Donald N. McCloskey (eds.), The Economic History of Britain since 1700, vol. 1 (Cambridge, 1994), 96-122.
Antràs, Pol, and Hans-Joachim Voth, “Factor prices and productivity growth during the English industrial revolution”, Explorations in Economic History, vol. 40 (2003), 52-77.
Baines, Edward, History of the cotton manufacture in Great Britain (London: Fisher, Fisher and Jackson, 1835).
Barton, D. Bradford, A history of tin mining and smelting in Cornwall (Truro: D.
Bradford Barton Ltd., 1967).
Brunt, Liam, “The advent of the sample survey in the social sciences”, Journal of the Royal Statistical Society, series D, 50 (part 2), 2001, 179-189.
Clark, Gregory and Jacks, David, Coal and the Industrial Revolution, 1700-1869, European Review of Economic History, Issue 1, (April 2007), pp 39-72.
Clinton, Alan, Post Office workers: a trade union and social history (London: George Allen & Unwin, 1984).
Cottrell, P.L., Industrial Finance, 1830-1914: The Finance and Organization of English Manufacturing Industry, London and New York (1980)
Crafts, Nick F. R., British economic growth during the Industrial Revolution (Oxford:
Clarendon, 1985).
Crafts, Nick F. R., and C. Knick Harley, “Output growth and the British industrial revolution: a restatement of the Crafts-Harley view”, Economic History Review, vol. 45 (1992), 703-30.
Dancy, Jeremiah, “British naval manning and its effects during the French Revolutionary Wars, 1793-1802”, unpub. D.Phil thesis, University of Oxford (2010).
Deane, Phyllis, and W. Arthur Cole, British economic growth, 1688-1959 (Cambridge:
University Press, 1962).
Ellison, Thomas, The cotton trade of Great Britain (London: 1886).
Flinn, Michael W., The British coal industry, vol. 2 (1700-1830) (Oxford: Clarendon Press, 1984).
Griffin, Emma, “A Short History of the British Industrial Revolution”, Palgrave Macmillan (November 2010).
Houston, Rab, Snell K.D., “Proto-industrialization? Cottage Industry, Social Change, and Industrial Revolution”, The Historical Journal, Volume 27, Issue 02 (1984), pp 473-492.
Hudson, Pat, “Regions and Industries: A Perspective on the Industrial Revolution in Britain”, Economic Geography, Vol 67, No. 2 (Apr., 1991), pp. 167-169.
Kitson, P. M., L. Shaw-Taylor, E. A. Wrigley, R. S. Davies, G. Newton, and M. Satchell,
“The creation of a ‘census’ of adult male employment for England and Wales for 1817”, unpub. mimeo., Cambridge (May, 2010).
Langton, J., “The Industrial Revolution and the regional geography of England”, Transactions of the Institute of British Geographers, ns 9, 145-67 (1984).
Lindert, Peter H., and Jeffrey G. Williamson, “Revising England’s social tables, 1688- 1812”, Explorations in Economic History, vol. 19 (1982), 385-408.
Marshall, Alfred, “Principles of Economics”, MacMilland and Co., London and New York (1890)
McCloskey, Deirdre, The Industrial Revolution, 1780-1860: A Survey, Chapter 8 , Economic History of Britain, 1700-Present, 2nd ed. (1994).
May, Trevor, An economic and social history of Britain, 1760-1970 (New York:
Longman, 1987).
Mokyr Joel, The British Industrial Revolution: An Economic Perspective (Westview Press, 2nd edition, 1993).
Moriss, Roger, The Royal dockyards during the Revolutionary and Napoleonic Wars (Leicester: University Press, 1983).
Porter, G. R., The progress of the nation (London: Methuen, 1912).
Schofield, Roger S., “British population change, 1700-1871”, in Floud, Roderick, and Donald McCloskey (eds.), The economic history of Britain since 1700
(Cambridge: University Press, 1994).
Shaw-Taylor, L., E. A. Wrigley, R. S. Davies, P. M. Kitson, G. Newton, and A. E. M.
Satchell, “The occupational structure of England c. 1710 to c. 1871”, unpub.
mimeo., Cambridge (April, 2010).
Shaw-Taylor, L., E. A. Wrigley, R. S. Davies, P. M. Kitson, G. Newton, and A. E. M.
Satchell, “The Creation of a ‘census’ of adult male employment for England and Wales for 1817”, unpub. mimeo., Cambridge (May, 2010).
Szostak, Rick, The Role of Transportation in the Industrial Revolution, (1991).
Trew, Alex, “Spatial takeoff in the first industrial revolution”, Review of Economic Dynamics, 17, 707-725 (2014).
Wood, George H., “The statistics of wages in the nineteenth century. Part 19 – the cotton industry”, Journal of the Royal Statistical Society, vol. 73 (4), 585-633 (1910).
Wrigley, E. A., “British population during the ‘long’ eighteenth century1680-1840, The Cambridge Economic History of Modern Britain, Voluem 1: Industrialisation 1700-1860, , p.57-96
Wrigley, Sir Tony, “The PST system of classifying occupations”, unpub. mimeo., Cambridge.
Trade directories as a data source on occupational structure: evidence from England in 1851
*Liam Brunt
†& Jan Erik Meidell
‡Abstract
In the absence of occupational census data before 1851, recent research utilizes baptism registers to infer occupational structure in England in 1817. As an alternative source, we propose using trade directories as these seem to be unaffected by many of the fundamental issues stemming from baptism registers. We outline the history of trade directories and detail their construction. Using the 1851 occupational census and contemporary trade directories, we show that it is possible to infer local occupational structure from trade directories with reasonable accuracy. A suitably stratified sample could generate a national occupational distribution. This technique could likely be employed back to 1770.
Keywords: Census, structural change, industrialization.
JEL classifications: J21, N13, O14, Y10.
* We would like to thank Lucy White and Jeff Williamson for helpful comments, as well as seminar participants at the London School of Economics, the University of Bergen, the Economic History Association Meetings in Boston and the Sound Economic History Workshop in Bergen. Any remaining errors are our own responsibility.
† Department of Economics, NHH – Norwegian School of Economics, Helleveien 30, 5045 Bergen, Norway; liam.brunt@nhh.no.
‡ Department of Economics, NHH – Norwegian School of Economics, Helleveien 30, 5045 Bergen, Norway; janerik.meidell@nhh.no.
0. Introduction. Recent characterizations of the British industrial revolution have played down the rate of economic growth, which is now widely agreed to have been slower than was suggested originally by Deane and Cole.1 Instead, more emphasis has been placed on the role of structural change, especially the transfer of labor resources from agriculture to industry.2 An obvious lacuna in this line of argument is that the available quantitative evidence on the rate or extent of structural change has been weak. The first census did not take place until 1801 and the occupation data that were collected in that year are worthless; households are categorized into three sectors (“Agriculture”, “Industry” or
“Other”) and for most counties these sum to something like 50 per cent of the number of households, leaving us to wonder what the rest of the population were doing. Only with the census of 1841 do we get the first reliable estimates of occupational structure; but the 1841 categorization remains primitive and, in any case, by 1841 the first stage of industrialization was almost complete, so those data are not much help in measuring structural change. There have been previous efforts to quantify English social structure in the eighteenth century3; these have formed the basis of important quantitative research.4 But social structure is not exactly the same thing as occupational structure (even though the two are linked); and the quantification has been fairly broad brush and based on very imperfect sources.
Recent research by the team at Cambridge led by Shaw-Taylor and Wrigley has used a variety of sources – particularly baptism records and militia ballots – to address this issue. They seem to paint a very different picture to the one to which we are accustomed, with much higher rates of industrialization by 1801.5 We discuss their research much more fully in the next section. Overall, we believe that there are significant challenges with their approach and that it is essential to consider the evidence of alternative sources. Trade directories contain large amounts of contemporary data on economic and employment structure and offer a potentially fruitful line of enquiry. Trade directories are a well-known source for historians but there seems to have been no attempt to harvest them for large-scale quantitative analysis. The purpose of this paper is to assess the possibilities of such a line of research and consider the difficulties that it poses. We therefore undertake a detailed comparison of contemporary trade directory data and the 1851 census – the first census for which there are really detailed occupational data. We show that trade directories do seem to offer a reliable guide to occupational structure, when handled with suitable care. This could perhaps push our knowledge of occupational structure back from 1851 to 1770.
In the next section we consider the recent research. In section two we outline the history of trade directories and consider why and how they were compiled. In section three we undertake the comparison of trade directories and the 1851 census. Section four concludes.
1 Deane and Cole, British economic growth; Crafts and Harley, “Output growth”; Antràs and Voth, “Factor prices”.
2 Crafts, British economic growth.
3 Lindert and Williamson, “Revising”.
4 Crafts, British economic growth.
5 Kitson et al., “Creation of a ‘census’”; Shaw-Taylor et al., “Occupational structure”; Wrigley, “PST system”.
1. Baptism records. From 1813 onwards, the registration of a baptism by the Church of England required the recording of the occupation of the father. For the period 1813-20, Kitson et al. took paternal occupation data from all extant registers (around 11 000) across England; for smaller parishes the population of registrations was entered, whereas sampling was used for larger parishes. These reported paternal occupations were used to infer occupational structure. Note that there are a number of sample selection issues here, some of which are discussed by Kitson et al.. Most obviously, the data pertain to males only, an important caveat that Kitson et al. state very clearly in their numerous papers.6 If males were employed systematically in different occupations – which seems highly likely – then the occupational distribution of males would not be a good guide to the overall occupational distribution. It is also problematic that many people were not Anglicans by this date – perhaps belonging instead to a dissenting Protestant church or the Catholic Church – and thus will not appear in the baptism sample. Various occupations – such as the law, the military and the Church of England – were off limits to people who were not members of the Established Church; therefore Dissenters must have had a different occupational distribution. There are several, rather more subtle, sample selection issues that we discuss below.
In table 1 we present the occupational structure estimates of Kitson et al., based on baptism registers, alongside those adopted by Crafts for 1801 and those available from the 1851 census. The Crafts data have been used repeatedly over the last 25 years as a basis for estimating economic growth; they show a very significant shift of labour resources out of the primary sector (agriculture, in fact) into the secondary sector. By contrast, the Kitson et al. data show a much higher share of industrial workers already by 1817, and a much lower share of service workers. This paints a very different picture of the rate of industrialization to that proposed by Crafts (and, later, Crafts and Harley).7 Kitson et al. find very little trace of industrialization in the early nineteenth century.
Instead they find a Commercial Revolution, with a dramatic relative shift of employment out of agriculture and into services. Of course, their estimates are not strictly directly comparable to the other estimates in table 1 because they are based on male occupations only; some kind of correction would need to be made to generate a consistent series, although we are not in a position to say what the nature of that correction would be.
Overall, we would say that either baptism registers must revolutionize the way that we think about British industrialization, or else we must have serious concerns about the validity of using them to infer occupational structure. We incline towards the latter point of view.
Table 1. Comparison of estimates of occupational structure.
1800 (Crafts)
1817 (Kitson et al.)
1851 (Census)
Primary 40 38 28
Secondary 30 42 41
Tertiary 30 19 32
Sources: 1800 – Crafts, British industrialization, 62; 1817 – Kitson et al., “Occupational structure”, 10.
6 Kitson et al., “Creation of a ‘census’”.
7 Crafts and Harley, “Output growth”.
Now we must ask what conditions Kitson et al. require for their inferences about occupational structure to be valid. There are two highly important ones.8 First, it must be the case that representatives of all occupations were equally likely to marry and procreate. This is clearly false. The likelihood of marriage was obviously related to income – paupers, prisoners and the laboring poor (for example) would be less likely to marry than industrial workers. But the situation is more complicated than this because there are both lifestyle and lifecycle considerations. It seems plausible that males in some occupations would be systematically less likely to marry and procreate – such as merchant seamen who were away on long voyages, members of the Catholic clergy and faculty members of Oxford and Cambridge. Inferring the proportion of these men in the population on the basis of the frequency with which they appear in baptism registers seems hazardous because their lifestyle makes procreation less likely. Consider also the case of soldiers and Royal Navy personnel. Since they were abroad for years on end, one might conclude that they are in the same situation as merchant seaman. But this is rather a problem of lifecycle than of lifestyle. Most soldiers were young; most fathers were older. It may be the case that many of the males in the baptism register sample were formerly soldiers; but they are unlikely to be soldiers at the point in time when we observe them having children. Therefore occupations that were disproportionately filled by young people – who later moved on to another type of work – will be systematically under-represented in the baptism registers. Agricultural and other labourers occupies a significant share of the samples in Kitson et al and thus represents an important source for errors. Note, in passing, that all the examples we have listed fall within the service sector – this will be of some importance in the following discussion.
There are also problems with multiple occupations. Many people had multiple occupations in the early nineteenth century and the question arises as to which occupation a father would offer if only one were recorded. If undertakers were commonly carpenters as well, then one could imagine that there may well be a systematic preference for reporting carpenter in preference to undertaker (or wagoner in preference to night soil collector, and so on). Kitson et al. discuss the assumption of equal marriage and fertility rates at some length and try to correct for the fact that it is violated, for example by adding extra laborers to their estimates. But their corrections are rather ad hoc and clearly extremely incomplete.
Second, Kitson et al. require that male completed fertility (that is, the total number of children produced per man) was the same for men across all occupations. We find this second condition – which Kitson et al. assume to be true – to be strong, surprising and unnecessary for the following reasons. It is a strong assumption in the sense that small violations have a large impact on the estimated occupational structure.
Suppose that we sample four baptism records and find that two fathers are recorded as agriculturalists and two as industrial workers. Kitson et al. then infer that there were two agricultural workers and two industrial workers in the workforce (an industrial share of 50 per cent). Now suppose that each agricultural worker had one child and each industrial worker had two children in their lifetime. Then, of course, there would really be two agriculturalists but only one industrial worker in the workforce (an industrial share of 33 per cent). As we sketch in table 2 below, small variations in completed fertility across occupations could change Kitson et al.’s estimates of occupational structure drastically.
8 As they note – Kitson et al., “Occupational ‘census’”, 3-4.
Kitson et al. observe the data in column 2 of table 2 and assume a one-to-one mapping to the number of fathers, thus generating column 3 and therefore column 4. Now suppose instead that the completed fertility of service sector families (such as merchant seamen and soldiers) were one third lower than average, and the completed fertility of industrial families one third higher, as inscribed in column 5. Then the observations in column 2 map instead to the number of fathers in column 6, and then on to the occupational structure of column 7. But columns 7 and 4 in table 2 are simply a restatement of columns 2 and 3 in table 1. Thus differential fertility rates could easily transform the estimates of Kitson et al. into the estimates of Crafts. In reality, we do not need to make such an extreme assumption as a one third difference in sectoral fertility rates. If we admit that Crafts’ estimates may be slightly out; and that there may have been some change already between 1800 and 1817; and if we allow for the fact that Kitson et al. consider only males; then a smaller assumed differential in fertility would be able to reconcile the two sets of numbers. But this need not lead to wholesale revisions of Crafts’
position, nor of his description of the pattern of industrialization.
Table 2. Possible effect of variations in occupational fertility.
1 2 3 4 5 6 7
Observed Baptisms
Inferred Number of
Men (Kitson et al.)
Inferred Occupational
Structure (Kitson et al.)
Hypothesized Occupation-
Specific Fertility
Inferred Number of
Men
Inferred Occupational
Structure
Primary 1.15 1.15 38 0.95 1.21 40
Secondary 1.25 1.25 42 1.35 0.93 30
Tertiary 0.60 0.60 20 0.65 0.92 30
Sources: see text.
We have now explained why we find the assumption of no variation in fertility across occupations to be strong. We find it also surprising because our understanding of the literature on English population growth is that a significant part of it was due to the movement of the workforce into industry. Changes in population growth between 1716 and 1816 were driven by changes in fertility; and changes in fertility were driven by changes in nuptiality, particularly the decline in the average marriage age of women by three years.9 Earlier marriage led to higher completed fertility (more children per woman) because the number of births per women was largely determined by the number of years of marriage of fertile women. Industrial workers married earlier because they reached their peak earnings at a younger age; thus industrial workers had higher completed fertility than agricultural workers and the move into industry accelerated population growth. If this story were true then surely we should expect to see higher completed fertility for industrial workers in 1801 also? If we would accept this assumption, that would explain why Kits et al conclude with a secondary employment of 42% in 1817.
This is higher than in 1851 (41%) and a significant higher level than Crafts estimation of 30% in 1801. In such circumstances, the contrary assumption – that completed fertility was the same across occupations – would not be innocuous.
We are also puzzled as to why Kitson et al. need to make this assumption at all.
They entered all the available baptismal records. Since the name of the father is noted on
9 Schofield, “British population change”, 73-81.
the record, they could simply count the total number of children produced by each man in the sample. Since they know also his occupation, they could work out total fertility rates for each occupation. There are certain statistical complications involved in this exercise but – given the large size of their sample – they should be able to get precise estimates.
This is an important issue to be resolved.
In general, Kitson et al. work at quite a high level of aggregation. That is, their results are presented for large geographical areas; occupations are highly aggregated, from 14 570 into only 113 occupations; and they pool data from many years. They note that some degree of aggregation is necessary because baptisms occur sufficiently infrequently that this generates small sample problems. This seems to be a significant drawback of using baptism records – there are not enough data to provide a really detailed picture.
Overall, we do not find any inexplicable inconsistencies between Crafts’ data and those of Kitson et al.. Plausibly adjusting their data for differential fertility rates across occupations, the absence of women, and other sample selection issues, could reveal two estimates of occupational structure that are very similar. Since the two estimates anyway pertain to benchmark years that are 16 years apart, we certainly could not say that the two estimates are substantially different. Whether they are statistically significantly different is also impossible to say, since Kitson et al. do not provide confidence intervals for any of their estimates.
We stress that the primary purpose of this paper is not to provide a detailed critique of the research of the Cambridge team; we have not examined their sources and methods in sufficient detail to pass judgment. Rather, we seek to make the following points in our discussion of their research. First, the headline figures produced by Kitson et al. seem quite at variance with Crafts, whose figures were previously generally accepted. Second, the differential could be due largely to (unmeasured) differential fertility rates, the exclusion of women and sample selection. Third, an alternative data source exists that is abundant, not subject to these biases, and has not previously been employed – namely, trade directories. Therefore, it seems sensible to see if the trade directory data can be harnessed to generate reliable estimates of occupational structure.
That is the purpose of this paper.
2. History of trade directories. Samuel Lee prepared the first British trade directory in 1677, but the entries covered only 1 953 wholesale merchants living in London.10 It seems to have met with limited success, since the exercise was not repeated until Henry Kent produced a new directory in 1734. Kent followed the same format as Lee but included 693 fewer names – so either London had shrunk or Kent’s directory was very incomplete, the latter seeming more plausible. Coverage seems to have improved over the first few editions (up to 2 006 entries in 1740) but Kent’s ambitions remained very limited in his subsequent annual revisions. Osborn’s London directory first appeared in 1740 and offered a wider range of information, but was seemingly still very incomplete.
The bar was raised in 1763 with the appearance of Mortimer’s Universal Directory. He included not only the merchants and bankers of London but also people in other trades and professions: artists, musicians, doctors, lawyers, booksellers, shopkeepers and so on.
10 This paragraph is based on Goss, London directories, 1-35.
By the early nineteenth century, the Post Office directory, which first appeared in 1800, was listing around 11 000 entries; and Johnstone’s 1817 directory was up to 27 000.
Importantly, Sketchley produced a directory for Birmingham in 1763 – the first for a town outside London.11 The first two editions of Sketchley’s directory have not survived but the third edition (1767) has a format very similar to Mortimer’s Universal Directory for London. Directories soon appeared for many other towns around England and thus trade directories could potentially constitute a very useful quantitative source on English economic history from 1763 onwards. Between 1763 and 1790, up to 50 new directories were produced. These covered ten towns and some also attempted to cover larger areas, with county directories appearing for Hampshire (1784) and Bedfordshire (1785). William Bailey, in 1784, was the first to attempt a national directory that covered the principal towns throughout the kingdom. Wilke’s Universal British Directory, which appeared in eight volumes between 1791 and 1798, raised the bar again by including also many smaller towns.
In the early nineteenth century, town and county directories became common. In total, Norton’s exhaustive survey counts 878 provincial (i.e. non-London) directories published before 1856. Many of these directories are readily available in electronic format because they are of interest to geneologists; therefore they constitute one of the most accessible historical sources. Over time, directories became more thorough and complete and were produced to a higher standard. Famous names – such as Pigot’s and White’s –started to appear in the 1810s; they set out to cover the whole country both systematically and repeatedly. From the perspective of the economic historian, repetition is a key ingredient. First, it may enable us to trace changes over time using a consistent source. Second, it probably generates a more accurate directory. How does repetition increase accuracy? The directory producer had an extra incentive to ensure that his directory was accurate because he had a reputation to maintain to generate future sales.
He also had experience of producing directories and thereby a better idea of how to elicit accurate information (as we discuss further below). Finally, the directory producer already had local knowledge when preparing his directory (i.e. the data base generated by the previous edition).
The issue of accuracy is, of course, crucial in historical enquiry. First consider what we mean by accuracy. It is obviously not the case that the entire population was listed in a trade directory. Poor people would not have been listed; nor would many better off people who were not involved in trade (for example, retired people or military officers or noblemen). In fact, it is highly unlikely that even all the traders were recorded.
There may be systematic omissions – such as dung collectors, who might not have wanted to advertise their trade – as well as random omissions and errors. In that sense, the directories are incomplete. But this does not make the directories useless. If we want to track accurately economic changes over time, or map variations across the country, then we do not necessarily need a complete register of all traders and producers. What we desire is transparency and, preferably, consistency. If we know the likely sources of error, so that we can correct for them; and if we know that these remained fairly constant over time; then we may be able to say something worthwhile about changes or variation in economic structure.
11 This paragraph is based on Norton, Guide, 1-15.
So how did directory producers compile their data? Several approaches seem to have been adopted.12 Early producers, such as Bailey and Pye, claim to have visited every house in the locality to elicit information from the householder. Pye, in fact, states that he gave up this approach in his later directories because it was too expensive, which sounds plausible. It may also have been counterproductive because people knocking unexpectedly at the door and asking about the nature of the householder’s business might be suspected of being tax collectors – and therefore lied to, or told to go away. In any case, personal interview could not have been a practical mode of compiling county or national directories because the task was simply too vast for a private entrepreneur. Thus it became common to use local agents to collect information.
How could the directory producer ensure that the data he received from local agents were accurate? Wilkes partially solved this problem in the Universal British Directory by enlisting local printers and booksellers as his agents; since they were then remunerated in the form of offprints for local sale, they had a stake in generating an accurate product. Logically, the first thing that a potential purchaser would examine to gauge the accuracy of a national directory would be his own town: if it were accurate then he might be willing to believe that the rest of the directory were accurate; if it were not then it would be difficult for the local bookseller to persuade him otherwise. Thus each local bookseller was likely to be able to retail his free offprints of the national directory only if he did a good job of collecting the data in his own town.
Another innovative approach was crowdsourcing. A draft of the local directory was left with a prominent resident of the town and people were asked to inspect and correct it; perhaps this is where Jimmy Wales got the idea for Wikipedia. Traders and professional people had an obvious incentive to ensure that the information about them was accurate and up to date. Not only might this attract business from out of town but one could also imagine that there was a certain cachet derived from being in the directory. In the nineteenth century a class of professional directory agents evolved, who presumably varied in quality – and needed to be monitored – just as other professional workers did (and still do today).
The overall impression is that the quality and completeness of trade directories varied enormously. This probably depended on the integrity of the producer and may also have depended on the target audience. For example, Wilkes’ Universal British Directory contains a short (or sometimes rather long!) description of the setting and history of each town. This would be particularly useful for someone travelling who wanted to plot an interesting route. These people were then likely to observe firsthand the quality of the directory because they would see how well the directory matched reality. By contrast, the Post Office London Directory was intended primarily for people sending letters, which might or might not arrive. A letter might go astray – or elicit no reply – for many reasons;
or the postman might manage to deliver the letter even with a wrong address; so the user of the directory had no obvious way of assessing the accuracy of the directory.13
12 These are discussed in Norton, Guide, 16-18.
13 In the early nineteenth century the cost of postage was paid by the recipient, rather than the sender, so postmen were highly motivated to ensure that the letter reached its destination – even if the address were not accurate. Payment by the recipient obviously gave the postman the correct incentives to handle the mail carefully and promptly; otherwise postmen would have had an incentive simply to store bags of undelivered mail in their lofts, as they do today.
Thus considerable caution is warranted when using trade directories as a source.
Consistency is probably more easily attained than accuracy. Depending on the historical issue being addressed, consistency may be a sufficient characteristic to ensure valid inference. Thus using the same directory for one place over time (such as multiple editions of Pigot’s or White’s), or using one directory that covered the whole country at a certain point in time (such as Wilke’s Universal British Directory) are likely to be the two most trustworthy strategies. It also seems likely that earlier directories will be less accurate and complete than later directories – there was less competition to provide directories, a smaller database on which to build, less experience of eliciting accurate information and less infrastructure on which to rely (such as professional directory agents).
3. Testing trade directories against the census. The structure of the historical problem that we need to solve is sketched in table 3 below. We would like to be able to track occupational change over time using the census but there was no occupational census before 1851. We would therefore like to create a synthetic occupational census for earlier years using some other source. Since we have trade directories for 1770 and 1851 – and, indeed, at numerous intermediate dates – they are a potentially valuable source if we could harness them correctly.
Table 3. Data sources available to track occupational change.
1770 1851
? Occupational census
Trade directory Trade directory
Is it possible to move from trade directories to an occupational census with a sufficient degree of accuracy to make the exercise worthwhile? What are the difficulties that we face? The first problem is that trade directories tell us about the number of businesses operating in each occupation, not the number of workers employed. We will therefore need to multiply each business by an employment factor that is appropriate to that occupation. Note, however, that this procedure is strength as well as a weakness.
When we multiply by employees per business, we implicitly include women as well men and those not affiliated to the Church of England.
The second problem is that the likelihood of a business appearing in the trade directory might be a function of its occupation. For example, it is plausible that businesses dealing directly with consumers (say, tailors) made sure that they were listed in the directory to obtain essential publicity, whereas businesses dealing with other businesses (say, ironworks) could successfully establish a reputation by word of mouth. If this were true then – even if we knew the average number of employees for each type of business – we would still not be able to estimate accurately the occupational structure of the population because we would have the wrong distribution of businesses across occupations.
We can lay these fears to rest using matched occupation and trade directory data from 1851. Logically, it should be possible to interpret the 1851 census as an enormous and complete trade directory for Great Britain. How? The 1851 census contains a table of
employees per business, broken down by occupation.14 Dividing the total number of people in each occupation by the average number of employees per business (in that occupation) should give the number of businesses in each occupation. That is, it forms a sort of national trade directory for Great Britain (albeit a trade directory with the street addresses and names of the businesses removed, which anyway are of no interest to us at this point). Census data are reported for each county and also major towns, as well as nationally; so it can be matched to town-level trade directories the we get a much finer geographical coverage than is possible with baptism records.
Of course, the procedure turns out to be rather more complicated than this. First, the 1851 table of employees per business enumerates only those businessmen (“Masters”) who have more than zero employees (“Journeymen and Apprentices”). So we have to infer how many businessmen there were who had zero employees. In principle, this is straightforward because, for each occupation, the table reports the number of employers having a particular number of workers. If we were to multiply all the employers in an occupation by the number of workers that each of them employed, then we should get the total number of people working in that occupation except those businessmen who employed zero. We could then compare this number to the total number of people recorded in the census as having that occupation. Any difference should (in theory) be composed of businessmen who had zero employees. The first problem with this exercise is that the number of employees is given only within certain bounds (1, 2, 3,… 10-19, 20- 29,… 50- 74,… 75-100,… 350 and over). We address this problem by assuming that – on average – each firm was located mid-way between its particular set of bounds. For example, we assume that firms in the 10-19 category employed 15 workers; this is the most plausible assumption and – in expectation – will minimize the magnitude of any error.
The second problem is that most occupations have a very large discrepancy between the two estimates of total workers (i.e. the estimated number of workers employed is much lower than that enumerated in the census). This implies that many occupations had an implausibly high frequency of businessmen who employed zero workers. For example, in order to reconcile the two estimates of the number of people working as bakers, it would have to be the case that 75 per cent of bakers employed no workers. It is possible that 75 per cent of bakers employed no help, but it is not the most plausible suggestion. The census therefore seems to be internally inconsistent.
An explanation for such inconsistency is offered on p. cclxxvi of the 1851 census itself. Many employers neglected to complete the part of the form asking about the number of their employees. This would lead us to incorrectly assume that all the missing bakers (who were not recorded as employees) were sole proprietors with no employees.
This would lead us to overestimate the total number of bakery businesses in Great Britain. For example, if a baker employed three people but neglected to note this in his census return then those three people would end up be counted as three one-man bakery businesses in our calculations. This could make it impossible for us to match the census with trade directories accurately.
We could therefore make one of two extreme assumptions. Either all the missing people in an occupation were one-man businesses; or all the businesses in that particular occupation employed people in the same size distribution that we observe in the table (i.e.
14 See British Government, Census of Great Britain, 1851: Population Tables II, vol. 1, cclxxvi-cclxxix.
